In recent years, with the rapid development of tiny electronic components. At present, commonly used optical inspection technology and electron microscope inspection technology find it difficult to meet the dimensional inspection requirements of key components of highly integrated integrated circuits. Based on computer micro-vision technology, the coupling of measuring platform displacement to the development of miniature electronic components is of great significance for the analysis and design of precision platforms. This paper analyzes the influence of the three factors of material, software and environment on the accuracy positioning of the measurement system. The analysis shows that the characteristics of the imaging system itself and the image noise it brings are the main factors that affect the measurement accuracy. In this paper, a technique for blindly detecting the position of the scanner based on computer vision is proposed. This article uses a CCD optical camera to capture the orbital motion of the electronic pen, and uses video image processing algorithms, computer vision and position to capture the position of the light on the screen. With the support of specific software, this article creates a new interactive whiteboard system. Any large screen can be turned into a large touch screen with electronic writing surface and touch function. The device does not require a specific additional screen to operate, and the cost is lower. In this paper, the stereo microscope has a narrow field of view, small depth of field and many non-linear factors. Based on the analysis of existing camera models, a binocular microscope imaging model based on a linear camera is established. On this basis, through the experimental analysis of the feasibility of the existing camera calibration method in the binocular microsystem, the calibration method of the camera's main parameters is proposed. Experiments show that the resolution value of the peak value of the evaluation curve in this paper is 18 times the focus resolution value, while the commonly used algorithm for evaluating the peak value of the curve is only 14 times. Compared with the latter, the discrimination of the edge gradient Laplacian used in this paper is more obvious than the former.